Don't undervalue the epeen, I loved my 64gb of ram
That is a factor, I have done that myself over the years sadly, most recently putting in 2 NVMe drives in RAID 0 for an extra like 300-400MB/sec of sequential read with most of the rest of the benefit being lost the the bottleneck of the PCI 4x bandwidth. Partly because the cost difference was very very small, but also partly because it's a bit of an epeen thing. Is there enough epeenery to justify a 2nd product line and all the overheads of engineering that go into that and the resulting more expensive card? I dunno, maybe. I suspect not.
Which isn't necessarily true, is it.
If you cache assets you don't need to use them any more than if you loaded everything "just in time". It just means they're available to be used when needed, without delay and without needing to be streamed from a much slower medium.
It's not always true but it's generally true. Bit for bit what you load into vRAM can have significantly different impacts on performance. You can add 1 new texture @ a few MBs and see very minor performance impact, or you could load a few kb pixel shader which does something extremely complex and halfs your frame rate. What is useful is talking about this in aggregate, you only need look at the evolution of GPUs over the last 20+ years to see that as GPUs get faster we add more vRAM.
I did acknowledge what you're saying, in my prior post, where I talked about vRAM being mostly necessary for what you're actually rendering right around you at the moment and that most games engines prefetch and cache assets they know or predict will be needed in future. Although that's really not as big of a deal as you'd imagine. People are correct in saying the vRAM is a way faster source of data than having to fetch from disk, but also 10Gb of vRAM (of which only some smaller subset is actually used for raw game assets) is not a lot of data, you can fill that from a modern disk super fast, relative to human perception.
Caching large volumes of assets you're not using right now is a bit of a waste of vRAM, it's a solution to a problem that costs money because vRAM costs money. Where as software can intelligently use small amounts of cache in an effective and efficient way to achieve basically the same thing, which is what game engines have been perfecting for 20+ years. My GTA V install is 92Gb which when uncompressed in memory after loading would be way more, and my 8Gb video card handles that game just fine, I can zip around the island in a jet and it has zero problems seamlessly streaming assets in/out intelligently.
I'm talking broadly here, what people do in reality is they get a game, they crank up the settings, they check their frame rate, if it's too choppy they lower the settings which lower vRAM usage on aggregate until they get a playable frame rate. If you have a 20Gb card and you lower your setting to be playable and at those settings you're using 10Gb of vRAM then you're wasting the other 10. There is a relationship between the GPU and vRAM and how much vRAM is appropriate for any given GPU, above which you're only adding cost and not really any actual benefit.